Multi-agent Reinforcement Learning in Sequential Social Dilemmas — mentioned by hcheemskerk

Sat, 08 Jun 2019 00:47:38 GMT

Sounds like you could adapt Leibo’s sequential social dilemma games. https://arxiv.org/abs/1702.03037 Instead of a total collective reward being the goal, you’d have team-based scores. You’d need cooperation within the team, and aggressive action against the enemy team. Here’s an implementation of 2 games https://github.com/eugenevinitsky/sequential_social_dilemma_...

Paper and discussion history. A mention is not an endorsement.

Multi-agent Reinforcement Learning in Sequential Social Dilemmas · HackerPapers discussions

Multi-agent Reinforcement Learning in Sequential Social Dilemmas — mentioned by hcheemskerk